Markers and visualization #622
Conversation
|
This can fail because the |
|
Well spotted! I'll think about how best to deal with this issue |
|
And the markers need to be cleared in |
|
@JLiesenborgs Are you currently working on how to deal with the issue still or were you able to solve it? |
|
I haven't had much time yet, but I do intend to look into this. Should I hurry? |
|
Im currently working on application of stella_Vslam marker integration as well. Just wanted to know more about your approach |
|
If we could connect over linkedin or email, I wanted to bounce some ideas off of you. |
|
@alvodo I apologize for taking this long, but I've finally added two commits to this pull request. The first should take care of the race condition: it records in which cases the vtx info was actually created. The second one is the small change to the 'clear' routine in map_database, as requested. If this looks acceptable, and you'd like to approve the pull request, it's probably best to squash the commit with the race condition fix into an earlier commit. I left it as a separate one for now for clarity, better suited for review. @tvt15 if you check out my fork of the repository, you can find an email address you can use in the commit log |
|
@JLiesenborgs I have sent you an email |
|
@JLiesenborgs No problem, thanks for the fix! I couldnt reproduce the problem with your new version, I would therefore conclude that it is fixed. I am not a maintainer or anything of this repo, so my approval wouldnt do anything. Maybe this is a good point in time for @ymd-stella to look at it? |
Loading/saving marker info, 2D and 3D, to sqlite3 keep_fixed_ flag in marker: previously, the marker positions were not optimized, the position was estimated once based on a single observation and kept fixed. The default is now to include the position in the optimization, which could still be turned off by setting the flag to true (would currently need to be done in the code) initialized_before_ flag in marker and "required_keyframes_for_marker_initialization" setting: if a single observation of the marker is used to estimate its 3D position, a bad estimate can still throw off the optimization procedure, causing the markers to mess up the map building instead of helping. The setting allows the initialization to wait until a number of keyframes have seen the marker. Only when this happens, the flag is set, and the marker will be included in the optimization. The default is to wait for 3 frames, but can be controlled using a config file setting.
This patch add support for using the aruco nano header file (https://sourceforge.net/projects/aruco/files/ArucoNano-v5/) to detect markers. The user can enable the support using a CMake config setting, and then use "aruconano" as marker model.
This draws the 2D detected markers, together with their IDs, on the observed image. Because this needs the original coordinates (not the undistorted ones), they are stored in the marker2d class now as well (not saved/loaded to database though). To allow a publisher (eg the socket_publisher) to share the 3D marker info, a method has been added to retrieve them.
JLiesenborgs
force-pushed
the
markers_and_visualization
branch
from
c5234f6 to
12e5d36
Compare
|
I've fixed these formatting issues (I ran the tool locally, so I hope I caught the same issues), and also squashed that fix for the race condition into the relevant commit. |
|
@alvodo @ymd-stella Just a small message to notify you - I forgot to mention you a few days ago, I don't know if you get to see the update without it (apologies if you were already informed) |
|
@JLiesenborgs Hey Jori, I sent you an email regarding your work. Please do check it out when you get the time. |
ymd-stella
force-pushed
the
markers_and_visualization
branch
3 times, most recently
from
5fc2425 to
f4ad896
Compare
- Delete references from markers when erasing keyframes - required_keyframes_for_marker_initialization can now be set individusually in KeyframeInseter and Initialiser
ymd-stella
force-pushed
the
markers_and_visualization
branch
from
f4ad896 to
d91beef
Compare
We have been experimenting with markers in stella_vslam. I cleaned up the modifications we made that seem to improve their use. I've tried to be quite elaborate in the commit messages, to explain what they do.